Applications of Discrete-time Markov Chains and Poisson Processes to Air Pollution Modeling and Studies (SpringerBriefs in Mathematics)

I + 1}. as a result, from (2. four) the posterior distribution of the parameter θ = (K, P(K) ) is given through (see [14]) ⎛ ⎡ ⎢ P(K, P(K) | Y) ∝ ⎣ I+1 ⎜ Γ ∑i=1 αmi ⎝ I+1 ∏i=1 Γ (αmi ) (K) ∏ m∈ χ2 I ∏ i=1 (K) × I 1 − ∑ Pmi (K) (K) (K) nmi +αmi −1 Pmi nmI+1 +αmI+1 −1 i=1 ⎞⎤ ⎟⎥ λ ok ⎠⎦ okay! (see additionally [19, 20]), and the marginal conditional posterior distribution of P(K) given ok is ⎧ (K) ⎪ (K) ⎨ Γ ∑I+1 I i=1 [nmi + αmi ] (K) nmi +αmi −1 (K) P P(P | ok, Y) = ∏ ∏ mi ⎪ I+1 Γ n(K) + α (K) ⎩ ∏ i=1 mi mi i=1 m∈ χ 2 (K) × I 1 − ∑ Pmi i=1 (K) ⎫ nmI+1 +αmI+1 −1 ⎪ ⎬ ⎪ ⎭ , (K) (2. 6) (K) i. e. , it's the manufactured from Dirichlet distributions with parameters nmi + αmi , m ∈ χ2 , i ∈ {1, 2, . . . , I + 1}. comment. observe that the chance functionality L(Y | okay, P(K) ) is proportional to a made of multinomial distributions and, by means of assumption, now we have that the earlier distribution P(P(K) | okay) is a fabricated from Dirichlet distributions. for this reason we have now that the posterior distribution P(P(K) | ok, Y) is usually a manufactured from Dirichlet distributions. this is often so as the set of Dirichlet past distributions varieties a conjugate kin of distributions with recognize to the multinomial probability functionality (see for example [23] and [34]). bear in mind that the process followed this is to estimate first the order of the Markov chain and afterwards estimate its transition possibilities. for this reason, we use first the mode of the marginal posterior distribution, P(K | Y), of the order okay, to estimate ok after which use the mode of P(P(K) | ok, Y) to estimate the transition percentages. with a view to do that we'll use the utmost a posteriori technique: hence we use, as estimates of the parameters, the values that maximize the respective posterior distributions (i. e. , the modes). The expression for P(P(K) | ok, Y) is given by means of (2. 6), and because the marginal chance functionality is (see [14] and [19]) 16 2 Markov Chain types L(Y | okay) ∝ ∏ (K) m∈ χ2 ⎧ ⎨ Γ ∑I+1 i=1 αmi I+1 ⎩ Γ ∑I+1 [n(K) + α ] mi i=1 mi ∏ i=1 ⎫ (K) Γ (nmi + αmi ) ⎬ , ⎭ Γ (αmi ) (2. 7) the marginal posterior distribution of ok is the discrete distribution given via ⎛ 1⎜ P(K | Y) = ⎝ c ∏ ⎧ ⎨ ⎩Γ (K) m∈ χ2 Γ ∑I+1 i=1 αmi (K) ∑I+1 i=1 [nmi + αmi ] I+1 ∏ i=1 (K) Γ (nmi + αmi ) Γ (αmi ) ⎫⎞ ⎬ λK ⎟ , (2. eight) ⎠ ⎭ ok! the place ⎛ c= ⎧ ⎨ ⎜ ∑⎝ ∏ k∈S (k) m∈ χ2 ⎩Γ Γ ∑I+1 i=1 αmi I+1 (k) ∑i=1 [nmi + αmi ] I+1 ∏ i=1 (k) Γ (nmi + αmi ) Γ (αmi ) ⎫⎞ ⎬ λk ⎟ ⎠ ⎭ okay! is the normalizing consistent. when we have the price of okay that maximizes (2. 8), the worth that maximizes (2. 6) is (k) Pmi = (k) nmi + αmi − 1 (k) ∑I+1 j=1 (nm j + αm j − 1) , i = 1, 2, . . . , I + 1, (k) m ∈ χ2 (2. nine) (k) (see for instance [30]). this is often so simply because, given ok = ok, the mode of P(Pm | ok = (k) okay, Y), m ∈ χ2 , is given through (2. 9). consequently, if we have to recognize what the likelihood of getting a size belonging to an period [L j , L j+1 ), is j = zero, 1, . . . , I +1 (taking L0 = zero and LI+2 = ∞), we in basic terms need to recognize the order of the chain and what the current country is, and we then use the transition matrix given through (2.